Draft version April 12, 2021 Typeset using LATEX twocolumn style in AASTeX63

SOFIA FEEDBACK survey: exploring the dynamics of the stellar wind driven shell of RCW 49

M. Tiwari,1, 2 R. Karim,1 M. W. Pound,1 M. Wolfire,1 A. Jacob,2 C. Buchbender,3 R. Gusten,¨ 2 C. Guevara,3 R.D. Higgins,3 S. Kabanovic,3 C. Pabst,4 O. Ricken,2 N. Schneider,3 R. Simon,3 J. Stutzki,3 and A. G. G. M. Tielens1, 4

1University of Maryland, Department of Astronomy, College Park, MD 20742-2421, USA 2Max-Planck Institute for Radioastronomy, Auf dem H¨ugel,53121 Bonn, Germany 3I. Physik. Institut, University of Cologne, Z¨ulpicherStr. 77, 50937 Cologne, Germany 4Leiden Observatory, Leiden University, PO Box 9513, 2300 RA Leiden, Netherlands.

(Received December 8, 2020; Accepted April 8, 2021) Submitted to ApJ

ABSTRACT We unveil the stellar wind driven shell of the luminous massive star-forming region of RCW 49 using 12 SOFIA FEEDBACK observations of the [C II] 158 µm line. The complementary dataset of the CO and 13CO J = 3 → 2 transitions is observed by the APEX telescope and probes the dense gas toward RCW 49. Using the spatial and spectral resolution provided by the SOFIA and APEX telescopes, we disentangle the shell from a complex set of individual components of gas centered around RCW 49. We find that the shell of radius ∼ 6 pc is expanding at a velocity of 13 km s−1 toward the observer. Comparing our observed data with the ancillary data at X-Ray, infrared, sub-millimeter and radio wavelengths, we investigate the morphology of the region. The shell has a well defined eastern arc, while the western side is blown open and is venting plasma further into the west. Though the stellar cluster, which is ∼ 2 Myr old gave rise to the shell, it only gained momentum relatively recently as we calculate the shell’s expansion lifetime ∼ 0.27 Myr, making the Wolf-Rayet star WR20a a likely candidate responsible for the shell’s re-acceleration.

Keywords: ISM: clouds — ISM: kinematics and dynamics

1. INTRODUCTION (2006) to report that these features are ubiquitous in One of the most important problems in modern as- our Galaxy and the authors coined the term “bubbles” trophysics is to understand the role of massive stars in by stating “We postulate that the rings are projec- driving various physical and chemical processes in the tions of three-dimensional shells and henceforth refer interstellar medium (ISM). Massive stars inject an im- to them as bubbles”. Processes that cause these shells mense amount of mechanical and radiative energy into can disrupt molecular clouds, thereby halting star for- their immediate vicinity. Stellar winds are responsible mation or can compress the gas at the edges of the H II for the mechanical energy input, which can push the regions, triggering star formation (Elmegreen & Lada gas into shell-like structures (as in the Rosette Nebula, 1977, Williams & McKee 1997 and Zavagno et al. 2010). arXiv:2104.04276v1 [astro-ph.GA] 9 Apr 2021 Wareing et al. 2018 and in the Orion Nebula, Pabst Thus, shells are ideal laboratories to study positive and et al. 2019). The radiative energy input comes from the negative feedback generated by massive star formation. heating of gas through stellar extreme-ultraviolet (EUV, Furthermore, in order to trace these shells, we can use + hν > 13.6 eV) and far-UV (FUV, 6 < hν < 13.6 eV) the 1.9 THz fine-structure line of ionised carbon, C photons that can ionize atoms, dissociate molecules ([C II] ), which is one of the major coolants of the ISM and also among the brightest lines in PDRs (Dalgarno and heat the gas giving rise to H II regions and pho- todissociation regions (PDRs). These stellar feedback & McCray 1972, Stacey et al. 1991, Bennett et al. 1994). As the ionization potential (11.3 eV) of carbon (C) is mechanisms power the expansion of H II regions and shock fronts causing morphological features that appear less than that of hydrogen (H) (13.6 eV), [C II] traces + as shells or bubbles in the ISM. Observational studies the transition from H to H and H2 (Hollenbach & Tie- at near-infrared (IR) wavelengths led Churchwell et al. lens 1999). While the [C II] layer probes warm atomic 2 Tiwari et al.

−1 2 Figure 1. Top panel: velocity integrated intensity maps in the range of -25 to 30 km s , from left to right: of [C II] P3/2 → 2 12 13 P1/2 fine-structure line, J = 3 → 2 transitions of CO and CO. The maps are in original resolution with their original beam sizes and the receiver array geometry is shown in the bottom left of each panel. Bottom panel: Emission images, from left to right, of the 8 µm GLIMPSE, 70 µm PACS, 870 µm ATLASGAL and ACIS data toward RCW 49. The center of Wd2 cluster at R.A.(α, J2000) = 10h24m11s.57 and Dec. (δ, J2000) = −57◦46042.500 is marked with a white asterisk. The OV5 star and the WR20b star are marked with yellow and pink asterisks, respectively. For [C II] emission, the contour levels are smoothed to a pixel size of 1500 and are 20% to 100% in steps of 20% of the corresponding peak emission. For the presented maps, the peak 12 13 −1 −1 −1 12 emission for [C II], CO and CO are 650 K km s , 300 K km s and 80 K km s , respectively. For CO emission, the contour levels are 10% to 100% in steps of 20% of the corresponding peak emission. For 13CO and 870 µm emission, the contour levels are 15% to 100% in steps of 20% of the corresponding peak emission. For 8 µm emission, the contour levels are 15% to 100% in steps of 30% of the corresponding peak emission. For 70 µm, the contour levels are 20% to 100% in steps of 30% of the corresponding peak emission. For 0.5–7 keV emission, the contour levels are 10% to 100% in steps of 30% of the corresponding peak emission. RCW 49 3 gas from the surface of clouds at low visual extinction FEEDBACK1 (Schneider et al. 2020) performed with (Av / 4 magnitude), the rotational transitions of CO the SOFIA and the Atacama Pathfinder Experiment 2 probe cooler molecular gas at larger Av (Hollenbach & (APEX , G¨ustenet al. 2006). The FEEDBACK Legacy Tielens 1999) deeper into the cloud clumps. Program was initiated to quantify the mechanical and radiative feedback of massive stars on their environment. RCW 49 is among the most luminous and massive star A wide range of sources were selected to be observed forming regions of the southern Galaxy located close that allow a systematic survey of the effects of differ- to the tangent of the Carina arm at l = 284.3◦, b = ent feedback mechanisms due to star formation activity -0.3◦. The earlier heliocentric distance measurements to (with a single O type star, groups of O type stars, com- RCW 49 varied from 2 to 8 kpc, as discussed in Church- pact clusters, mini starbursts, etc.), morphology of their well et al.(2004), Rauw et al.(2007), and Furukawa environment, and the evolutionary stage of star forma- et al.(2009). Drew et al.(2018), in their discussion of tion. In particular, RCW 49 was selected to study the the distance, note that photometric studies of the stars feedback of the compact stellar cluster, Wd2 and the powering RCW 49 have more recently tended towards Wolf-Rayet stars on their surrounding molecular clouds. 4–6 kpc rather than the 2–8 kpc span that is usually We use the fine-structure line of [C II] to probe the quoted, although Rauw et al.(2007, 2011) maintain that shell associated with RCW 49 and disentangle its dense the distance must be 8 kpc. Three of the most recent gas component using the CO observations. We quantify works determine a distance of about 4.2 kpc (Vargas the stellar wind feedback responsible for the evolution of Alvarez´ et al. 2013; Zeidler et al. 2015; Cantat-Gaudin the shell of RCW 49 and describe the shell’s morphology. et al. 2018). Vargas Alvarez´ et al.(2013) and Zeidler In Sect. 2, we describe the observations. The qualitative et al.(2015) in particular conducted photometric stud- and quantitative analysis of the data are reported in ies with two independent sets of observations and agreed Sect. 3. Both small and large scale effects of the stellar on a ∼4 kpc distance. This is consistent with the 4.2 kpc feedback in RCW 49 are discussed in Sect. 4 and the Gaia parallax distance reported by Cantat-Gaudin et al. results of this study are summarised in Sect. 5. (2018), though Drew et al.(2018) point out that the uncertainties on these small parallax values are consid- 2. OBSERVATIONS erable. In accordance with the photometric study of 2.1. SOFIA Observations ´ Vargas Alvarez et al.(2013) and the consistent mea- The [C II] line at 1.9 THz was observed during three surement by Zeidler et al.(2015), we adopt a distance flights from Christchurch, New Zealand on 7th, 10th, of 4.16 kpc. and 11th of June 2019, using upGREAT3 (Risacher RCW 49 contains a bright H II region ionized by a et al. 2018). upGREAT consists of a 2 × 7 pixel low- compact stellar cluster, Westerlund 2 (Wd2), compris- frequency array (LFA) that was tuned to the [C II] line, ing 37 OB stars and ∼ 30 early type OB star candidates and in parallel a seven pixel high frequency array (HFA) around it (Tsujimoto et al. 2007; Ascenso et al. 2007; that was tuned to the [O I] 63 µm line. Both arrays ob- Rauw et al. 2011; Mohr-Smith et al. 2015; Zeidler et al. serve in parallel, but here, we only present the [C II] 2015). There is a binary Wolf-Rayet star (WR20a) as- data. The half-power beam widths are 14.100 (1.9 THz) sociated with the central Wd2 cluster, suggested to be and 6.300 (4.7 THz), determined by the instrument and one of the most massive binaries in the Galaxy (Rauw telescope optics, and confirmed by observations of plan- 00 et al. 2005). RCW 49 also hosts an O5V star and an- ets. The final pixel size of the [C II] map is 7.5 . The other Wolf-Rayet star (WR20b), both a few arcminutes observation region was split into 12 individual ‘tiles’, away from the geometrical cluster center. These and a each covering an area of (7.26 arcmin)2 = 52.7 arcmin2. handful of other massive stars in the cluster periphery During the three flights, eight tiles were observed (∼66% may have been ejected from Wd2 (Drew et al. 2018). of the planned area). Each tile was covered four times Age estimates generally suggest that the cluster is not much older than 2 Myr (Ascenso et al. 2007; Zeidler et al. 2015). 1 https://feedback.astro.umd.edu 2 APEX, the Atacama Pathfinder Experiment is a collaboration between the Max-Planck-Institut f¨urRadioastronomie, Onsala In this paper, we report one of the first results Space Observatory (OSO), and the European Southern Observa- of the Stratospheric Observatory For Infrared As- tory (ESO). tronomy (SOFIA, Young et al. 2012) legacy program 3 German Receiver for Astronomy at Terahertz. (up)GREAT is a development by the MPI f¨ur Radioastronomie and the KOSMA/Universit¨atzu K¨oln,in cooperation with the DLR In- stitut f¨urOptische Sensorsysteme. 4 Tiwari et al. CII 8 µm 870 µm CII 8 µm 870 µm

1 to 9 km s-1 northern cloud (F09)

North North East East

11 to 21 km s-1 cloud (F09) Inner dust ring (C04)

Ridge Transition boundary (C04) Shell

1 to 9 km s-1 southern cloud (F09)

Figure 2. RGB images of [C II] (red), GLIMPSE 8 µm (green) and ATLASGAL 870 µm (blue) emission toward RCW 49. The Wd2 cluster’s center, the OV5 and the WR20b stars are marked with white, yellow and pink asterisks, respectively. The ridge and the shell are marked in the left panel, while the inner dust ring (white dashed circle) and the transition boundary (white dotted line) are marked similar to Churchwell et al.(2004, Fig. 1) in the right panel. The 12CO clouds as shown in Furukawa et al.(2009, Fig. 1 (a) and (c)) are also outlined. In yellow are the northern and southern blobs of the cloud within the velocity range of 1 to 9 km s−1 and in magenta is the cloud within the velocity range of 11 to 21 km s−1. These clouds are discussed more in Sections 3.2, 3.3 and 3.4. and they were tilted 40◦ against the R.A. axis (counter center was at 10h24m11s.57, -57◦4604200.5 (J2000), the clockwise against North) for horizontal scans and per- reference position at 10h27m17s.42 -57◦1304200.60. For pendicular to that for the corresponding vertical scans. more observational and technical details, see Schneider As a consequence of its hexagonal geometry, the ar- et al.(2020). ray was rotated by 19◦ against the tile scan direction to achieve equal spacing between the on-the-fly scan As backend, a Fast Fourier Transform Spectrometer lines. The gaps between the pixels are approximately (FFTS) with 4 GHz instantaneous bandwidth and a fre- two beam widths (31.700 for the LFA and 13.800 for the quency resolution of 0.244 MHz was used (Klein et al. HFA), which result into a projected pixel spacing of 2012). The [C II] data thus have a native velocity resolu- 10.400 for the LFA and 4.600 for the HFA after rotation tion of 0.04 km s−1. We use here data that was re-binned (for more details, see the SOFIA Science Center’s Plan- to a resolution of 0.2 km s−1. Spectra are presented on 4 ning observations webpage ). The second two coverages a main beam brightness temperature scale Tmb with an are then shifted by 3600 to achieve the best possible average main beam efficiency of 0.65. The forward effi- coverage for the [O I] line. All observations were carried ciency is ηf = 0.97. From the spectra, a first order base- out in the array-on-the-fly mapping mode. The map line was removed and the data-quality was improved by identifying and correcting systematic baseline features with a novel method for data reduction that makes use 4 https://www.sofia.usra.edu/science/proposing-and- observing/observers-handbook-cycle-9/6-great/62-planning- of a Principal Component Analysis (PCA) of reference observations spectra as described in AppendixA. RCW 49 5

−1 −1 Figure 3. Velocity channel maps of [C II] emission toward RCW 49 with a channel width of 2 km s . The velocity (in km s ) of each channel is shown in top left of each panel. The Wd2 cluster’s center, the OV5 and the WR20b stars are marked with white, yellow and pink asterisks respectively. 2.2. APEX Observations 130◦ for the orthogonal scans, respectively. The map RCW 49 was mapped on September 25-26, 2019 in center and the reference position were the same as the good weather conditions (precipitable water vapor, pwv SOFIA observations. The latter was verified to be free = 0.5 to 1 mm) in the 13CO(3-2) and 12CO(3–2) tran- of CO emission at a level of < 0.1 K. A total area of 2 sitions using the LAsMA array on the APEX telescope 570 arcmin was observed, split into 4 tiles. Each tile 00 (G¨ustenet al. 2006). LAsMA is a 7-pixel single polariza- was scanned with 6 spacing (oversampled) in scanning 00 tion heterodyne array that allows simultaneous observa- direction, with a spacing of 9 between rows, resulting in tions of the two isotopomers in the upper (12CO) and uniformly sampled maps with high fidelity. All spectra lower (13CO) sideband of the receiver, respectively. The are calibrated in Tmb (main-beam efficiency ηmb = 0.68 array is arranged in a hexagonal configuration around a at 345.8 GHz). A linear baseline was removed, all data −1 central pixel with a spacing of about two beam widths resampled into 0.2 km s spectral bins. The final data 00 (θ = 18.200 at 345.8 GHz) between the pixels. It uses cubes are constructed with a pixel size of 9.5 (the beam mb 00 a K mirror as de-rotator. The backends are advanced after gridding is 20 ). FFTS (Klein et al. 2012) with a bandwidth of 2 × 4 GHz 2.3. Ancillary data and a native spectral resolution of 61 kHz. The mapping was done in total power on-the-fly mode, We present the images of the most relevant ancillary with the scanning directions against N at -40◦ and - data toward RCW 49 in the lower panel of Fig.1. We started with the Galactic Legacy Infrared Mid-Plane 6 Tiwari et al.

Figure 4. Velocity channel maps of 12CO (3-2) emission toward RCW 49 with a channel width of 2 km s−1. The velocity (in km s−1) of each channel is shown in top left of each panel. The Wd2 cluster’s center, the OV5 and the WR20b stars are marked with white, yellow and pink asterisks respectively. Survey Extraordinaire (GLIMPSE, Benjamin et al. instrument of the 12 m APEX telescope. The 870 µm 2003) 8 µm data observed with the Spitzer Space Tele- traces the cold and dense clumps shielded from FUV ra- scope. The 8 µm emission arises from the PDR sur- diation by large amounts of dust extinction. Hence, this face of dense molecular clouds where large hydrocar- map provides a good probe of the earliest star formation bon molecules, the Polycyclic Aromatic Hydrocarbons sites. Lastly, we used the archival data from the Chan- (PAHs), are excited by strong UV radiation resulting dra X-ray Observatory using its primary camera, the in fluorescent IR emission (Tielens 2008). Next, we Advanced CCD Imaging Spectrometer (ACIS) (Garmire used the 70 µm data from the Herschel Space Archive et al. 2003). Diffuse X-ray (0.5–7 keV) structures sur- (HSA), obtained within the Hi-GAL Galactic plane sur- rounding massive star-forming regions have been shown vey (Molinari et al. 2010) observed the with Photodetec- to trace hot plasmas from massive star feedback (e.g. tor Array Camera and Spectrometer (PACS, Poglitsch Townsley et al. 2019). et al. 2010) aboard the Herschel Space Observatory (Pil- 3. RESULTS bratt et al. 2010). The 70 µm emission traces the warm interstellar dust exposed to FUV radiation from the 3.1. Multi-wavelength overview of RCW 49 stars. Further, we obtained the 870 µm dust continuum In order to investigate the relation between the atomic data from APEX Telescope Large Area Survey of the and molecular components of the gas, we show the emis- Galaxy (ATLASGAL, Schuller et al. 2009), performed sion toward RCW 49 in different transitions and contin- with the Large APEX BOlometer CAmera (LABOCA) uum wavelengths in Fig.1. RCW 49 7

Figure 5. Velocity channel maps of 13CO (3-2) emission toward RCW 49 with a channel width of 2 km s−1. The velocity (in km s−1) of each channel is shown in top left of each panel. The Wd2 cluster’s center, the OV5 and the WR20b stars are marked with white, yellow and pink asterisks respectively. clouds (−11 to 9 and 11 to 21 km s−1) in RCW 49 and suggested their collision (∼ 4 Myr ago) to be respon- sible for triggering the formation of Wd2. The cloud within −11 to 9 km s−1 has a mass of (8.1 ± 3.7) × 4 10 M and extends over a range of ∼ 26.1 pc in the north-south direction and ∼ 21.6 pc in the east-west di- rection. The cloud within 11 to 21 km s−1 has a mass 4 of (9.1 ± 4.1) × 10 M and extends over a range of ∼ 12 13 Figure 6. Average spectra of [C II], CO and CO to- 18.3 pc in the north-south direction and ∼ 21.6 pc in the ward the whole mapped region of RCW 49. To highlight the east-west direction (Furukawa et al. 2009). The −11 to structures seen in the velocity channel maps, we mark the 9 km s−1 cloud further includes two seemingly different boundaries of the shell’s expansion (in green), the northern clouds: −11 to 0 km s−1 and 1 to 9 km s−1 (Furukawa and southern clouds (in orange) and the ridge (in pink). The et al. 2009, Fig 1 (c) and (d)), but were analysed as one. velocity integrated intensity maps of these regions are shown This is perhaps due to the unavailability of high spatial in Fig.7. resolution in the large scale 12CO data. We manually −1 −1 12 outline the 1 to 9 km s and 11 to 21 km s clouds A study of the large-scale CO (2 → 1) distribu- in the right panel of Fig.2. The −11 to 0 km s−1 cloud tion by Furukawa et al.(2009) identified two molecular will be discussed in detail in the next sections of this pa- 8 Tiwari et al.

12 13 Figure 7. Left to right: Velocity integrated intensity maps of [C II], CO and CO of the shell’s expansion within -12 to 0 km s−1 (top row), the northern and southern clouds within 2 to 8 km s−1 (middle row) and the ridge within 16 to 22 km s−1 (bottom row). The Wd2 cluster’s center, the OV5 and the WR20b stars are marked with white, yellow and pink asterisks respectively. RCW 49 9 per. Churchwell et al.(2004) studied RCW 49 in mid-IR structures in the various maps reveals the complex me- wavelengths to investigate its dust emission morphology chanical and radiative interaction of the Wd2 cluster and identified distinct regions as a function of the angu- with the surrounding molecular gas. In this paper, we lar radius with respect to Wd2. It can be seen in Fig.1 will use the kinematic information in the [C II] and CO that all tracers, except the X-ray emission, are devoid emission maps to focus on the kinetics and energetics of any emission in the immediate surrounding of Wd2. of the arc-like emission structure. In a future paper, we This emission free region is filled by hot plasma (tem- will examine the other emission components present in perature of ∼ 3 × 106 K and density of ∼ 0.7 cm−3, see this data. Sect. 3.5.2) as evident from its very bright X-ray emis- sion (as seen in Fig.1). Moving away from Wd2, the 3.2. Velocity channel maps 8 and 70 µm emission becomes brighter and marks the 12 13 so called “transition boundary” (at ∼ 5 pc from Wd2, A spatial comparison of the [C II], CO and CO shown in Fig2, right panel), a ring-like structure opened emission in different velocity channels can be seen in to the west (Churchwell et al. 2004, Fig. 1). Dense cores Figs.3,4 and5. In the [C II] channel maps (Fig.3), we are traced by ATLASGAL 870 µm. A dense ridge struc- can see that a shell like structure starts to develop from −1 ture (pointed in Fig.2, left panel), running from south ∼ -24 km s and it appears to be expanding in the −1 to east of Wd2, is particularly prominent in the AT- velocity range of ∼ -12 to 0 km s . A red blue image LASGAL emission map. This ridge is also visible in the is shown to depict this expansion in the AppendixB, 8 µm PAH and 70 µm dust continuum emission but not Fig. 14. The eastern arc of the shell is well defined com- very prominent in the CO (3 → 2) emission maps. It pared to the western arc, which appears to be broken. −1 is important to verify the spatial and spectral informa- In the velocity range of 2 to 12 km s , we see [C II] tion given by 12CO observations with 13CO because in emission confined to the north and south. The ridge as general, 12CO is optically thick and could be impacted we discussed in Sect. 3.1 is seen in the velocity range of −1 by opacity effects causing self absorption, etc. Thus, 16 to 22 km s . 12 13 the optically thin 13CO is used to confirm the results Channel maps of CO and CO (Figs.4 and5) show obtained from 12CO observations. In addition, the com- similar velocity structures. The shell seen in the [C II] bination of 12CO and 13CO can be used to determine channel maps is also outlined by fragmented emission 12 13 the molecular gas mass. In contrast, the dense cores to from both CO and CO, starting from about -12 to −1 the west and north are also recognizable in CO (3 → 0 km s . Similar to the [C II] emission, the eastern side 2) emission maps but not so prominent in 8 µm and of the shell is more apparent than its western counter- 70 µm emission. Our 12CO (3 → 2) map follows the part. The northern and southern structures are spread −1 larger scale 12CO (2 → 1) emission distribution as re- out over a smaller velocity range (2 to 8 km s ) com- ported in Furukawa et al.(2009). In addition to the pared to the [C II] emission. Though the ridge is not above mentioned bright emission, both have a slightly very distinct in CO emission, it appears to be associated 12 fainter emission toward the west of Wd2, which is absent with the molecular cloud traced by CO for velocities −1 in the ATLASGAL emission. greater than 16 km s . The [C II] emission map shows similarities to these 3.3. Different structures in RCW 49 structures. It reveals a lack of emission immediately 12 Figure6 shows the average spectra of [C II], CO and surrounding Wd2. One of the most prominent struc- 13CO toward all of the mapped regions shown in Fig1. tures in the [C II] map is a bright and wide arc (labelled We mark the boundaries of three main structures that “shell” in Fig.2, left panel) of emission running from the we identify in the channel maps (Figs.3,4 and5): the east to the south which is quite prominent in the channel shell’s expansion, the northern and southern clouds and maps from -14 to -6 km s−1 (see section 3.2). At higher the ridge. Figure7 shows these structures as seen in II velocities the [C ] emission is more and more domi- 12 velocity integrated intensity maps of [C II], CO and nated by the ridge southeast from Wd2 and a separate, 13CO. The top row of Fig.7 shows the integrated inten- dense structure to the north. The brighter [C II] peaks sity of the velocity range within which the expansion of in these morphological structures have counterparts in the shell is clearly visible (in Fig.3) and we discuss this the CO maps. We notice that the ridge structure is shell in detail in the further sections. The northern and much less bright in the [C II] line than in the 8, 70, and southern clouds and the ridge are shown in the middle 870 µm maps. This can also be viewed in the red green and bottom rows. These two structures are similar to blue (RGB) image of [C II] , 8 µm and 870 µm emission the clouds as reported by Furukawa et al.(2009) and as shown in Fig.2. The difference in the behavior of these shown in Fig.2. 10 Tiwari et al.

3.4. Spectra toward different offsets In contrast to the blue-shifted velocity component, the −1 Figure8 (left and middle panels) shows examples of red-shifted velocity component (> 15 km s ) of the 12 13 observed spectra does not follow any obvious trend. It observed spectra of [C II], CO and CO toward dif- ferent offsets along the horizontal and vertical cuts as corresponds to the ridge as discussed in Sect. 3.3 and shown in the right panel. These cuts were chosen to vi- shown in Fig.7 bottom row. The velocity components 12 13 −1 sualize the spectral line profiles of the observed species of [C II], CO and CO that lie within 2 to 12 km s toward the shell, which we see in the velocity channel are part of the northern and southern clouds (discussed maps (in Figs.3,4 and5). in Sect. 3.3 and shown in Fig.7 middle panel). The 00 00 There are multiple velocity components toward any spectra shown in Fig.9 at ∆ α = 50 , ∆δ = -300 show 12 13 position but by examining closely the line profiles of CO and CO line profiles peaking between the two the blue-shifted velocity component, we can follow the peaks of [C II] line, which is indicative of self-absorption −1 shell’s progression. As we move along the horizontal cut at velocities > 0 km s . This spectral behaviour could 00 00 (at ∆δ = −10000) shown in the right panel of Fig.8, we also be seen at an offset of ∆α = 100 , ∆δ = -250 . This suggests that the northern and southern clouds probably see that the blue-shifted velocity component of the [C II] line sequentially starts to shift its peak from ∼ 0 km s−1 lie in front of the gas constituting the shell. (at ∆α = 30000) to ∼ −11 km s−1 (at ∆α = 10000) and back to ∼ −4 km s−1 (at ∆α = −20000). Thus, tracing 3.5. Expanding shell of RCW 49 12 the shell’s expansion. For the CO emission, the blue- Owing to the high spectral and spatial resolution of 12 13 shifted velocity component roughly follows the [C II] line our [C II], CO and CO data, we were able to dis- 00 profiles. It starts to show up at ∆α = 200 , sequentially entangle different components of gas along the same −1 00 shifting its peak to ∼ 14 km s (at ∆α = −100 ) and line-of-sight. The [C II] channel maps (Fig.3) and the −1 00 back to ∼ −4 km s (at ∆α = −200 ), similar to [C II] spectra displayed in Fig.8 reveal a blue-ward expand- 12 line. As mentioned earlier (in Sect. 3.1), CO is usually ing shell to the east of Wd2. The two (in red and blue) 13 optically thick, thus, we need to examine CO spectral [C II] velocity channel maps displayed in Fig. 10 b out- line profiles to confirm the presence of different velocity line this arc structure particularly well. The west side 12 components seen in CO spectra. The blue-shifted ve- somewhat mirrors this arc-like structure but it is clearly 13 locity component of CO emission line intensity is low a less coherent structure. The blue-shifted channel maps and is not detected with our S/N. However, it can be of the 12CO and 13CO emission reveal a highly frag- seen that toward the lines-of-sight where the red-shifted mented clumpy distribution coincident with this limb- 12 13 velocity component of CO is relatively brighter, CO brightened shell. The position-velocity (pv) diagrams 12 is detectable and follows CO’s line profile. Thus, we (Figs. 10 e, 15 and 16) also reveal clearly the eastern arc 13 expect CO emission line to have similar profiles for its as well as the fragmented western arc. The red-shifted blue-shifted velocity component as well. (velocities > 1 km s−1) counterpart of the east and west A similar trend is observed in the spectra reported arcs are not very prominent in the channel maps (Figs.3, 00 along the vertical cut (at ∆α = 100 ). A shift in the 4,5 and 10 d), but there is evidence for a large scale, peak of the blue-shifted velocity component of [C II] can though highly fragmented, arc-like structure in the red- −1 00 be seen starting from ∼ 0 km s (at ∆δ = −300 ) shifted velocities pv diagrams as well. Perusing the pv −1 00 to ∼ −13 km s (at ∆δ = 200 ) and then back to ∼ diagrams (Fig. 10 e), we discern a spheroidal shell struc- −1 00 00 −5 km s (at ∆δ = 400 ). At ∆δ = 100 , we see no ture as outlined by large dots in Fig. 10 b. Assuming −1 emission from the shell (< 0 km s ) as evident from that the shell is expanding isotropically, its projection both the spectrum and its spatial distribution described on position-velocity space will be an ellipse. The maxi- 12 13 in Sect. 3.5. The CO and CO emission also follow a mum observed velocity and radius of the shell depend on similar trend as the spectra along the horizontal cut. the cosine of the angle between the center of the shell In addition to the spectra shown in Fig.8, we selected and a given cut along which a pv diagram is consid- a few more positions specifically along the shell to exam- ered (for details see Butterfield 2018). The horizontal ine the spectral line profiles of the observed species. Fig- straight dashed line in the pv diagrams, represents the 12 13 ure9 shows the spectra of [C II], CO and CO at dif- systemic velocity ∼ 1 km s−1, estimated by examining ferent offsets marked on the shell as seen from the veloc- the velocity profiles of [C II] emission toward various po- −1 ity (-25 to 0 km s ) integrated intensity map of [C II]. sitions. The systemic velocity can vary by 1–2 km s−1 Most of the spectra show a prominent blue-shifted (< when looking at the spectra along different offsets. This −1 0 km s ) velocity component, which is tracing the shell. is also consistent with the 12CO data. The maximum ob- served velocity of the shell is ∼ 13-14 km s−1, estimated RCW 49 11

12 13 Figure 8. Spectra (left and middle panels) of [C II], CO and CO toward different offsets along the horizontal and vertical −1 lines shown on the [C II] velocity (-25 to 30 km s ) integrated intensity map in the right panel. The presented spectra are smoothed to a velocity resolution of 1 km s−1. The sequential shift in the peak of the blue-shifted component is also marked with vertical dashed lines in some panels. The blue-ward shift of the expanding shell is quite apparent when comparing these spectra (see Sections 3.2, 3.3 and 3.4). from the observed spectra along different horizontal and of the gas comprises dense clumps that are moving at vertical cuts through the shell. The center of the ex- higher speeds ∼ 20 km s−1. The red-shifted gas compo- panding shell is estimated from the intersection of the nent, which we link to the northern and southern cloud longest horizontal and vertical cuts at ∼ ∆α = 000, ∆δ structures and the ridge, will be discussed in a future = 5000, which is ∼ 10000 east of Wd2. The resulting pre- paper. dicted ellipse is shown (solid curve in Fig. 10 e) for our observed shell. Comparing the horizontal and vertical 3.6. Dynamics of the shell cuts and their corresponding pv diagrams (in Fig. 16), 3.6.1. Mass of the shell we get a vertical (north-south) radius of ∼ 7.5 pc and a horizontal (east-west) radius of ∼ 4 pc, such that the We have used several independent methods to esti- geometric mean elliptical radius will be ∼ 5.5 pc and a mate the mass of the expanding shell. Each of these thickness of ∼ 1 pc are estimated. Thus, the shell has methods relies on calculating the mass of the thin, expanded more in the north-south direction than the limb-brightened arc of emission as identified on the east-west direction. [C II] channel maps. We then use a geometric model to In column e of Fig. 10, we flipped the ellipse (shown in estimate the total mass of the shell. dashed curves) of our blue-shifted shell and found that the actual structures at these higher velocities are not In the first method, we used 70 and 160 µm data from traced well by the ellipses. Instead the red-shifted part Herschel PACS to estimate gas column densities via the dust column. Illuminated dust throughout high-mass 12 Tiwari et al.

12 13 Figure 9. Spectra of [C II] (black), CO (red) and CO (blue) toward different offsets along the shell, shown around the −1 velocity integrated (from -25 to 0 km s ) intensity map of [C II]. star forming regions is heated by the FUV radiation and the 70 and 160 µm bands in our dust emission spectrum re-emits this energy in the far infrared (FIR) (Hollen- analysis. Castellanos et al.(2014), in their Section 4.3, bach & Tielens 1999). make a comparable analysis of these FIR observations The thermal FIR emission spectrum can be modeled of RCW 49 using these two PACS bands. as a modified black-body spectrum with spectral index Following the general technique of Lombardi et al. β in order to parametrize the emission in terms of dust (2014), we zero-point calibrated the PACS images by (effective) temperature and optical depth. With predic- predicting the intensity in the PACS 70 and 160 µm tions from dust grain models, such as those of Draine bands at 50 resolution using the Planck GNILC fore- (2003), the derived optical depth can be converted to ground dust model (Planck Collaboration et al. 2016). a hydrogen nucleus column density. We are ultimately We compare the Planck-predicted emission to the ob- interested in deriving the mass in the shell, which is served Herschel emission in order to determine the lined with PDRs and thus contains warmer dust. The 70 image-wide correction for each band. Because we are and 160 µm PACS bands are more sensitive to warmer making a significant extrapolation in predicting the (T > 20 K) dust than the longer-wavelength SPIRE shorter-wavelength PACS intensities using the Planck bands (250, 350, and 500 µm), so we elect to use only dust model, which is based on longer-wavelength Planck RCW 49 13

12 13 Figure 10. The top, middle and bottom rows are for [C II], CO and CO, respectively. The maps shown here are smaller cutouts of the ones shown in Fig.1. Columns a and c show the velocity integrated intensity maps of the observed species in the velocity ranges of −25 to 0 km s−1 and 0 to 30 km s−1, respectively. Column b shows red blue (RB) velocity channel maps 12 13 −1 −1 of [C II], CO and CO ranges from −12 to −8 km s (blue) and from −8 to −4 km s (red), respectively. The shell is 12 marked with white dots in the three rows of column b. Similarly, column d shows RB velocity channel maps of [C II], CO and 13CO from 0 to 4 km s−1 (blue) and from 4 to 8 km s−1 (red), respectively. Column e shows the pv diagrams along the vertical white dashed line cuts in the columns a and c. The predicted ellipse is shown on the pv diagrams for the blue-shifted part (in solid curve) and it’s flipped (in dashed curve) to the red-shifted velocity structures. observations, we make this comparison under a mask ex- to each image. The 70 and 160 µm intensities along cluding the warm central region. This exclusion limits the limb-brightened shell are of the order ∼ 0.5 to 2 × the comparison to lines of sight with low temperature 104 MJy sr−1, so our applied zero-point corrections of variation compared to the central region and still in- 80 and 370 MJy sr−1, respectively, are no more than cludes a large area of ∼ 25 K dust (see Figure 11) which 10% of the total intensity in either band. is reasonably bright from the Planck wavelengths up to After zero-point correcting the two PACS images, we PACS 70 µm, maintaining the validity of our extrapo- derive dust effective temperature and optical depth. We lation. A Gaussian curve is fitted to the distribution can model the emission with a modified black-body spec- of differences between the predicted and observed in- trum using Equations 1–3 from Lombardi et al.(2014). tensity for each of the two PACS bands, and the fitted In order to compare model emission with the PACS in- mean is assigned as the required zero-point correction. tensity measurements, we use Equation 8 from Lombardi The correction, a single number for each band, is added et al.(2014) to integrate the model spectrum over the 14 Tiwari et al.

Figure 11. Map of temperatures (left) and log-10 column densities (right) derived pixel-by-pixel from the Herschel 70 and 160 µm data as described in Section 3.6.1. The inset images in the upper right corners zoom in on the central region and show the half-ellipse mask used to estimate the shell mass from dust, [C II], and CO(3−2). relative spectral response functions available for both a fixed spectral index β = 2, consistent with the grain the 70 and 160 µm PACS bands. In the present work, all models of Draine(2003). source intensities and response functions which we dis- Since we have two observations and two unknowns, we cuss refer to the extended emission versions (as opposed are able to derive a unique solution for effective temper- to point source emission) where applicable. The PACS ature T and optical depth τ0. The simplest solution can photometry is expressed as intensity (MJy/sr), already be found by making the optically thin (τν  1) approx- −τν accounting for beam areas. Following their Equation 8, imation (1 − e ) ≈ τν . Applying this approximation we can write the mean intensity Ii measured in band i within our Equation2, measured intensity Ii in either as band can be expressed R Iν Rν dν h i h
βi Ii = . (1) I = BP B (T ) τ = BP B (T ) τ ν/ν R (ν /ν) R dν i i ν ν i ν 0 0 i ν (4) h
βi Expressing the above function of Iν as a “bandpass = BPi Bν (T ) ν/ν0 τ0 function” BPi of the incident source intensity Iν and Recalling that the bandpass function BPi is primarily combining with their Equation 1, we rewrite Ii as an integral over frequency, the constant-in-frequency τ0 h i h i can be pulled outside of the function. The ratio of the −τν Ii = BPi Iν = BPi Bν (T )(1 − e ) (2) intensities in the two bands excludes the parameter τ0 entirely. where Bν (T ) is the Planck function in Equation 2 by Lombardi et al.(2014) and τν is modeled as a power law h
βi BP70 Bν (T ) ν/ν0 with spectral index β, as given in their Equation 3. I70 = (5) h βi I160 BP B (T ) ν/ν
β 160 ν 0 τ = τ ν/ν
. (3) ν 0 0 This expression for the ratio of the measured intensi- We adopt ν0 = 1874 GHz, corresponding to 160 µm, ties depends only on one parameter, the effective tem- so that τ0 is the optical depth at 160 µm (τ160). We use perature T . The expression is easily evaluated for a RCW 49 15 range of T , producing a series of modeled intensity ra- not in the dust. We find that the masses (estimated from tio values. Using this numerical grid, we interpolate 13CO using the same technique as explained in the later from the observed intensity ratio values to temperatures. paragraphs of this section) along the brightest lines-of- With derived effective temperatures in hand, we rear- sight in CO split roughly 60/40, shell/ridge. The mass range our Equation4 and evaluate the expression for τ0 corresponding to the ridge part comes out to be 297 M using the measured intensities in one of the bands. putting an additional < 5% error on the shell mass. Our gas mass estimate from the dust column also in- I τ = i (6) cludes more diffuse foreground and background contri- 0 h
βi BPi Bν (T ) ν/ν0 bution, as the observed FIR intensity, and consequently the calculated column density, is non-zero even a few We have numerically evaluated the above two expres- arcminutes away from the shell. This is distinct from sions to derive the temperature and dust optical depth the non-shell components discussed above in that this over the map. We have converted the calculated 160 µm diffuse contribution is larger scale, extending past the optical depth into H-nuclei column densities (described central RCW 49 region in Figure 11, and is probably below) and the results are shown in Fig. 11. In prin- physically distant from and thus completely unrelated ciple, there is no need to make the optically thin ap- to the shell. We make a rough estimate of this com- proximation. We can work directly from Equation2 bined “background” by sampling a column density of and write the intensity ratio of the two bands without 1.3×1022 cm−2 a few arcminutes away from the shell. any approximation or cancelling of terms. We can then If we subtract this background from the map and then use the calculated optical depth in the optically thin carry through our previous calculation, we find that it approximation to derive, from the rewritten intensity may account for up to 30% of the mass in our upper limit ratio, an improved temperature, and use that to numer- estimate. Rather than make this rough background es- ically derive a new optical depth from Equation2 and timate and subtraction, we simply reiterate our inter- continue this iteration until convergence is achieved. As pretation of the mass measurement as an upper limit. the 160 µm optical depth in the shell is rather small, this Finally, the half-ellipse model captures most of the procedure converges rapidly. Tests on a few points in the visible [C II] shell, but the shell extends slightly past the shell demonstrate that the iteration produces only small northern edge of the mask by about 20◦ in azimuth, and (∼5%) changes in the optical depth. As this is compa- past the southern edge by about 10◦. We make a rough rable to the calibration uncertainty, we have elected to correction for this by multiplying the mass estimate by continue the analysis with the optically thin approxima- 7/6, assuming a constant linear density along the edge tion. of the shell. It is possible to make a more articulated In order to convert the 160 µm optical depth to hydro- shell mask for a more precise measurement, but this gen nucleus column density, N(H) = N(HI) + 2N(H2), would necessitate more complex assumptions. we use the Draine(2003)R V = 3.1 value of the dust ex- tinction cross section per hydrogen nucleus at 160 µm, −25 2 The second method to estimate the gas mass is by de- Cext,160/H = 1.9 × 10 cm /H, and solve Equation7 termining the C+ column density N(C+). As detailed for N(H). The maps of dust temperature and N(H) are 13 in AppendixD, the [ C II] line shows that optical depth presented in Figure 11. effects are small over most of the [C II] arc, except for

τ160 = (Cext,160/H) × N(H) (7) the brightest emission spot. Hence, we have calculated the C+ column density in the optically thin limit follow- We assume a half-elliptical shell, with the major and ing Tiwari et al.(2018, equation A.1). We assume an minor axes lengths from Section 3.5, and create a mask excitation temperature ∼ 100 K (lower limit), which is tracing the limb-brightened Eastern edge. From the a reasonable kinetic gas temperature in the [C II] emit- H nucleus column densities, which range from 0.3 to ting layer of a PDR to excite the C+ from 2P → 1.3×1023 cm−2 along the shell, we calculate a total gas 1/2 2P . We calculated the column density (upper limit) mass (excluding He) of this region, finding a value of 3/2 of the observed limb-brightened part for a region similar 8.5×103 M . to the mask used in the dust mass estimation and for the We consider this mass estimate an upper limit due to emission within -25 to 0 km s−1. We found an average line-of-sight contribution. This measurement picks up N(C+) ∼ 7.2 × 1018 cm−2. The column density N(C+) some other components of gas that are not part of the is not very sensitive to the choice of T . For instance, shell. One of the brightest lines-of-sight in 12CO (east ex assuming T = 200 K instead of 100 K decreases the of Wd2) is a superposition of the shell and ridge compo- ex calculated N(C+) by 19%. nents, which can be disentangled in the CO emission but 16 Tiwari et al.

A pixel-by-pixel sum of N(C+) over the entire thick- we determined numerically. By a simple argument ness allowed us to estimate the H gas mass. Using the of symmetry, this correction is valid for our quarter- abundance ratio of C/H = 1.6 × 10−4 (Sofia et al. 2004), spheroid shell assumption. Applying that factor to our 5 3 we get the corresponding H gas mass ∼ 4.6 × 10 M , limb-brightened part’s mass (derived from dust), and which is very similar to the mass calculated from the including the corrective factor of 7/6 explained earlier, 4 dust emission. we get the corrected shell mass estimate ∼ 2.5×10 M .

Finally, we can estimate the H2 gas mass from We can check our assumption of the shell mass against N(13CO), which requires determination of its excita- observed extinction through the foreground gas and tion temperature, Tex. Again, these estimations were shell. If we assume the gas distribution is uniform done for the same masked region as described above and throughout the shell, then we can divide the total shell for the emission within -25 to 0 km s−1. As detailed in mass by the surface area of a quarter of a spheroidal shell AppendixE, we get an average Tex ∼ 14.5 ± 1 K and at its mean semimajor and semiminor axes a = 7 pc and 13 15 −2 an average N( CO) ∼ (8.8 ± 0.1) × 10 cm such b = 4 pc and then convert the surface mass density to AV 12 17 −2 21 −2 that the average N( CO) ∼ 4.6 × 10 cm , using using the factor N(H)/AV = 1.9 × 10 cm (Bohlin 12 13 CO/ CO = 52 (Milam et al. 2005). Similar to the et al. 1978) for an RV = 3.1 reddening law. Including calculation of H gas mass, by summing pixel-by-pixel the geometric factor of 2.5 (but excluding the 7/6 fac- 13 12 −5 the N( CO) and by using CO/H2 = 8.5 × 10 (Tie- tor so that we match our surface area assumption), the 3 lens 2010), we get an H2 gas mass of ∼ 1.5 × 10 M . mass estimate from far-infrared dust emission results in As expected from the more fragmented CO emission, an extinction of AV, shell ∼ 18 through the shell. this mass estimate is somewhat less than derived from Vargas Alvarez´ et al.(2013) and Mohr-Smith et al. the dust or the [C II] emission. (2015) both measured an average AV ∼ 6.5 towards Wd2 cluster members, each finding the reddening to be In summary, our mass estimate using the dust includes RV ∼ 3.8. Hur et al.(2015) reported an abnormally both the atomic and molecular gas. The estimate using high reddening law, RV = 4.14, towards early-type clus- [C II] emission (CO dark gas) depends on the molecular ter members and a more typical law, RV = 3.33, towards fraction but differs by a factor of only 1.5 between pure foreground stars in the same field. They report a total atomic and pure molecular columns. The estimate from E(B − V ) ≈ 1.7 to 1.75 for most cluster members, and 13 CO is for the molecular gas alone. Thus the shell find a foreground E(B − V )fg ≈ 1.05, which suggests 3 mass estimated from dust (8.5 × 10 M ) is compara- AV, fg ∼ 3.5. Zeidler et al.(2015) report E(B−V ) ≈ 1.8 13 ble to that estimated by [C II] and CO together (6.1 to 1.9 based on reddening of line emission from ionized 3 × 10 M ). The range of masses obtained through the gas. different methods is about a factor of 1.4 in mass. All four of these measurements are consistent with AV ∼ 6.5 towards Wd2 cluster members, implying In order to estimate the entire shell’s mass, we assume AV, shell ∼ 3 towards the cluster after accounting for the that the shell occupies the space between two concentric foreground extinction measured by Hur et al.(2015). prolate spheroidal shells with North-South semimajor This suggests a significantly thinner shell than the axes a = 6.5 and 7.5 pc and semiminor axes b = 3.5 and AV, shell ∼ 18 we predict from our shell mass estimate, 4.5 pc such that the shell thickness is about 1 pc. We assuming a uniform shell. We explain this discrepancy also assume that the limb-brightened region is repre- by proposing that the thickness of the shell varies sig- sented by the volume of a prolate spheroid of a = 7.5 pc nificantly across its surface, a claim further supported and b = 4.5 pc from which the volume of an elliptic by the fragmented CO distribution we observe across the cylinder of a = 6.5 pc and b = 3.5 pc is subtracted, shell and the higher AV ∼ 15 fit by Whitney et al.(2004) which we call a “cored spheroid”. The conversion from to two young stellar objects embedded in the western the volume of a cored spheroid to that of a spheroidal shell. If we picture an optical path originating from the shell gives us a geometric correction factor of 2.5, which cluster and extending east, passing through the bright eastern shell, this path experiences AV ∼ 18 extinc- tion according to our mass calculations. But accord- 5 We note that our analysis assumes a purely atomic hydrogen column density but the C+ arises in both the warm atomic and ing to the optical extinction measurements, the thick- molecular gas. Since the collisional excitation rate of C+ by H2 is ness of the shell drops dramatically as the optical path ∼ 0.7 times the rate of excitation by atomic hydrogen (Wiesenfeld sweeps towards us. In fact, we do not detect any bright & Goldsmith 2014) we expect that a purely H2 column would shell component along the line of sight towards the clus- have ∼ 1.5 times the mass. RCW 49 17 Table 1. Densities, temperatures and pressures calculated for the shell of RCW 49.

−3 −3 −3 −3 Region n (cm ) T (K) pth/k (cm K) prad/k (cm K) pturb/k (cm K) Plasmaa 0.71 3.13 × 106 4.9 × 106 -- Ionized gasb 317 7.7 × 103 4.9 × 106 -- c 3 6 6 6 PDR/[C II] layer 4 × 10 300 1.2 × 10 2.6 × 10 5.9 × 10 Notes: Columns from left to right are region, H density, temperature, and thermal, radiation, and turbulent pressures. a − n is abundance of e from ionization of H and pth/k = 2.2nT , where the factor 2.2 accounts for ionization of H and singly ionized He. b − + n is e density from ionization of H and pth/k = 2nT , where factor 2 accounts for ionized H , and neutral He. c 2 n is H density and pth/k = nT . The radiation pressure prad = Lbol/4πkR c, where R is the radius of the shell. 2 The turbulent pressure pturb = µmn ∆vturb/8 ln 2 k, where µ = 1.3 is the mean molecular weight, m is hydrogen 2 2 mass and ∆vturb = ∆vFWHM − (8 ln 2 kT/mc), where mc is the carbon mass. ter in the [C II] and CO spectra and pv diagrams. We The combined mass loss rate, mechanical energy injec- ˙ 2 ˙ draw the conclusion that the extinction associated with tion (1/2 Mv∞), and momentum transfer rate (Mv∞) the cluster probes a much thinner section of the shell by the O and B stars within 30 of the cluster cen- than the limb-brightened eastern shell from which we ter, at the peak of the X-ray emission (Townsley et al. −5 −1 extrapolate our mass estimate, and so our geometrically- 2019), is (3.1 ± 0.2) × 10 M yr , (8.3 ± 0.5) × extrapolated mass measurement must be considered an 1037 ergs s−1, and (5.6±0.4)×1029 dyn (Leitherer et al. upper limit. 2010). The WR binary WR20a contributes an addi- +0.25 −5 −1 +1.3 37 −1 One could imagine combining our limb-brightened tional 1.9−0.20 × 10 M yr , 3.6−1.1 × 10 ergs s , +0.70 29 shell mass estimate with an optical extinction map like and 3.0−0.60 ×10 dyn. Using the evolutionary spectral those shown in Figure 14 in the paper by Hur et al. synthesis software Starburst99 (Leitherer et al. 2014, (2015) or Figures 8, 9, and 25 in the paper by Zei- see description in AppendixG), we estimate that over dler et al.(2015) in order to understand the variation the lifetime of the cluster (∼2 Myr), the OB stars have in thickness across the shell. High confidence in the injected ∼ 6 × 1051 ergs via their winds (see Fig. 12 and 3-dimensional positions of the cluster members and the AppendixG for additional detail). geometry of the shell would be required in order for such WR stars represent a rather short-lived phase of the a measurement to have any meaning, and this is beyond lifetimes of very massive stars, forming after about ∼2 to the scope of the present paper. 3 Myr, depending on their mass, and lasting a few hun- dred thousand years if we use the results of Starburst99 3.6.2. Energetics as a guide. The components of WR20a are estimated by Rauw et al.(2005) to be ∼ 80 M each, the most Using the total mass (excluding He) of the shell ∼ 2.5 4 −1 massive observed stars in the cluster; if this is close to × 10 M and its expansion velocity ∼ 13 km s , we 49 their initial masses, then neglecting binary effects on calculated its kinetic energy, Ekin ∼ 4 × 10 ergs. their evolution, Starburst99 would suggest that they formed after ∼ 3 Myr, creating some tension with most To assess the contribution from stellar winds in driv- of the age estimates of the cluster. If these Wolf-Rayet ing the shell, we need to estimate the stellar wind energy stars originated as much more massive objects, like the of Wd2. The early-type cluster members are catalogued ∼ 126 M predicted by Ascenso et al.(2007) to be the along with their established or estimated stellar types most massive star in the cluster based on their assumed by Tsujimoto et al.(2007), Vargas Alvarez´ et al.(2013), IMF, then would suggest that WR stars and Mohr-Smith et al.(2015). We used the theoretical Starburst99 could have formed after ∼ 2 Myr, which agrees better calibrations of Martins et al.(2005) to estimate effec- with independent age estimates. tive temperature Teff, surface gravity log g, and lumi- Rauw et al.(2005) observe evidence of enhanced sur- nosity L from the spectral type. For WR20b (WN6ha; face hydrogen abundance in the components of WR20a, van der Hucht 2001) and each component of the WR20a indicating that they (and WR20b, if we assume the com- binary (WN6ha+WN6ha), we assume parameters fit- ponents are identical) are still in the core hydrogen- ted to WR20a by Rauw et al.(2005). See AppendixF burning phase and, based on their position in the for additional details about the synthesized catalog, the Hertzsprung-Russell diagram, are only ∼ 1.5 Myr old measurements derived from the catalog, and the uncer- and have present-day masses very similar to their initial tainties on those measurements. 18 Tiwari et al. masses. In any case, over the lifetime of the cluster, the geometry with an outer radius of 5 pc, bordering the OB stars will have dominated the kinetic feedback; but PDR, and an inner radius of 2.7 pc, bordering the X- over the last 2 to 3×105 years, the winds of the WR ray emitting plasma, and calculated the electron density binary alone will have contributed ∼ 4 × 1049 ergs. and temperature (listed in Table1). Given the similarity in spectral characteristics, the Lastly, for the PDR, we compared our observations contribution by WR20b is likely similar to that of a with existing PDR models (PDR Toolbox; Kaufman single component of WR20a (approximately half the et al. 2006, Pound & Wolfire 2008). To use these, we values given above). However, WR20b is significantly needed an estimation of the FUV flux incident on the 0 offset (> 3 ∼ 3.5 pc) from the center of the X-ray PDR. We used the Teff and log g (or appropriate WR emission tracing the ∼ 3 pc radius plasma bubble, so parameters from Rauw et al. 2005) for each catalogued it is probably not playing as direct a role as WR20a in star to select models from the PoWR stellar atmosphere powering this bubble. grids (Sander et al. 2015). From the synthetic spectra provided by the PoWR models, we integrated the total The importance of the plasma’s thermal energy in ex- flux between 6 to 13.6 eV. Using the stellar coordinates panding the shell can be assessed by comparing thermal and these FUV fluxes from all stars within 120 (includ- pressures, pth, in the hot plasma, ionized gas and PDR ing WR20a and WR20b, though these contribute only a of RCW 49. To characterize the hot plasma, we use few percent of the total FUV flux), we estimated the in- the RCW 49 Chandra/ACIS observations reported by tegrated FUV flux between 6 to 13.6 eV, often expressed 3 Townsley et al.(2019). These authors fit X-ray spectra as G0 in terms of the Habing field, to be ∼ 2–3 × 10 towards RCW 49 with plasma models using the spec- in Habing units at the limb-brightened shell radius of tral fitting software Xspec (Arnaud 1996). The plane- ∼ 5–6 pc from Wd2. This value should be considered parallel, constant temperature shocked plasma model la- an upper limit, as the extinction between the illuminat- beled pshock2 represents the diffuse plasma towards the ing cluster and the PDR due to dust in the H II region center of Wd2, so we take from this model the fitted has not been accounted for. For the average intensities temperature and surface emission measure listed in the (within the masked region shown in Fig. 11) of [C II] Table 5 by Townsley et al.(2019). They independently (∼ 112 K km s−1) and 12CO (∼ 41 K km s−1) at a 3 fit spectra from the inner region towards Wd2 and the FUV radiation field, G0 ∼ 10 in Habing units, we deter- outer region out to ∼3 arcminutes away (see their text mined the PDR’s H density and temperature using the for details), and label these independent fits “Wd2 in- models from the PDR toolbox6 (Kaufman et al. 2006; ner” and “Wd2 outer”. We take the pshock2 parameters Pound & Wolfire 2008). Using our observed line ratio 12 −1 of the “Wd2 outer” region and assume the plasma fills a of [C II]/ CO(3-2) (after conversion from K km s to sphere whose circular cross-section equals the observed erg cm−2 s−1 sr−1) and comparing with the modeled 7 areas of “Wd2 outer” and “Wd2 inner” combined, yield- line ratio as a function of the cloud density and G0, ing a radius of 2.7 pc. allowed us to estimate the density. Further, using this Following the calculations of Townsley et al.(2003), density and the G0, we constrained the PDR tempera- we obtain the electron density (assuming a line-of-sight ture using the modeled PDR surface temperature map8. distance 2r through the plasma) and temperature, which The temperature is not strongly dependent on G0 in the are listed in Table1. Finally, assuming the tempera- derived density range. ture and pressure are constant throughout the spherical A list of the derived densities, temperatures and ther- bubble, the total thermal energy of the plasma is about mal pressures is presented in Table1. 48 2.4 × 10 ergs. There is clearly a mismatch (lower) in Using the sum of the bolometric luminosities, LBol, the thermal energy of the hot plasma, the mechanical for the OB stars and the WR20a star, we can esti- energy injected over the lifetime of the stellar cluster, mate the radiation pressure (as mentioned in Table1). and the kinetic energy of the expanding shell. We will The bolometric luminosities of the OB stars within 30 revisit this in Sect. 4.1. were taken from the spectral type calibrations of Mar- We followed the work by Paladini et al.(2015) to esti- tins et al.(2005) as described above and Rauw et al. mate the temperature and density in the ionized gas in the H II region using their observed H109α line proper- 6 ties. We excluded from the calculation their “region B” http://dustem.astro.umd.edu/ 7 http://dustem.astro.umd.edu/models/wk2006/ciico32web.html due to its outlying line-of-sight velocity, which indicates 8 http://dustem.astro.umd.edu/models/wk2006/tsweb.html it may be from the blue-shifted foreground ridge compo- nent we detect in [C II]. We assumed a hollow spherical RCW 49 19

Figure 12. Starburst99 predictions of the mechanical luminosity (left panel) and momentum transfer rate (right panel) due to the stellar winds. The Starburst99 simulation configurations are described in AppendixG. The solid blue and dashed red lines, and associated solid shaded blue and hatched red regions, give the values for OB and WR stars, respectively. The horizontal lines, solid and dashed and their associated shaded regions, mark the values calculated from the observed OB and WR stars as described in Section 3.6.2 as well as in AppendixF. The darker-shaded regions reflect uncertainty in the total cluster mass, as described in the AppendixG. The lighter-shaded regions reflect total cluster mass uncertainty as well as uncertainty in the maximum stellar mass; the upper limit uses a [1, 120] M range, and the lower limit uses a [1, 80] M . Note the effect of the maximum stellar mass on the age at which WR stars appear in these models. (2005) provides the bolometric luminosities of the com- Kee 1977) may have led to regions that are dense and ponents of WR20a based on their fit to its spectrum. cool enough to allow rapid cooling, resulting in a rapid We can also estimate the turbulent pressure, pturb (in loss of thermal energy. Table1), from the full width half maximum of the ob- We also recognize that there is a timescale issue. served [C II] emission line profile. These results reveal The shell radius of 6 pc and the expansion velocity of that there is rough pressure equipartition between the 13 km s−1 imply an expansion timescale of ∼ 0.5 Myr. thermal, turbulent and radiation pressure in the PDR. For an enclosed bubble driven by adiabatic expansion of Examining Table1, we conclude that the hot plasma, the hot plasma created by a continuous input of mechan- the ionized gas and the PDR are in approximate pressure ical energy by stellar winds, the expansion time scale is equilibrium as expected for a stellar wind shell driven by only 0.27 Myr (using equations 51 and 52 of Weaver mechanical energy input from the central star cluster et al. 1977). In contrast, the age of the Wd2 cluster (Weaver et al. 1977). is ∼2 Myr according to most age estimates. It should be noted that a different choice of the heliocentric dis- 4. DISCUSSION tance (say up to 8 kpc) of RCW 49 would increase the 4.1. Morphology of the shell and the role of WR20a in estimated expansion timescale of the shell to 0.52 Myr, its expansion which is still inconsistent with the age of Wd2. Hence, As discussed in Sect. 3.3.2, the kinetic energy of the the average expansion velocity over most of the cluster −1 expanding shell is much higher than the thermal energy lifetime must have been / 2 km s and only very re- of the plasma. We emphasize that the mechanical lumi- cently (/ 0.2 Myr), the shell has been accelerated to −1 nosity of the stellar cluster well exceeds the requirements 13 km s . We infer that feedback from OB stars ini- for driving the shell. Therefore, we surmise that much tially drove shell formation and expansion but that this of the thermal energy was lost once the shell broke open bubble quickly burst, releasing the hot plasma. At that to the west and the hot gas expanded freely into the en- point, expansion would rapidly slow down due to contin- vironment (as shown in Fig. 13). Beside the adiabatic ued sweeping up of the cold gas in the environment. The cooling associated with this “free” expansion, evapora- recent re-acceleration of the shell might be connected to tion of entrained cold gas into the hot plasma due to the evolution of the most massive stars (WR20a and electron conduction (Weaver et al. 1977; Cowie & Mc- 20b) to the Wolf-Rayet phase. If we assume the bub- 20 Tiwari et al.

limb-brightened shell

observer

Wd2 transition boundary (C04)

C+ expanding shell -1 vexp = 13 km s

transition boundary (C04)

Figure 13. 2D (upper panel) and 3D (lower panel) representations of RCW 49’s shell as seen by the observer. The transition boundary from Churchwell et al.(2004, Fig. 1) is marked and labelled with ‘C04’. The plasma (in blue), ionized gas (in green) and the PDR (in red) are shown. The limb-brightened part of the shell in the 2d illustration is actually the expanding [C II] shell seen in the 3D diagram. The transition boundary overhangs from the [C II] shell into the ionized gas structure behind it. ble has burst, expansion must be driven by momentum 4.2. Previous studies and larger scale structure transfer. The issue is that the Wolf-Rayet stars do not Our picture of a single shell at the center of RCW 49 seem to inject significantly more momentum than the differs from that of the two shell (separated by the ridge) ensemble of OB stars; this issue arises in the momen- scenario presented by Whiteoak & Uchida(1997) and tum transfer rates calculated directly from the observed Benaglia et al.(2013). Owing to the kinematic informa- WR stars and their properties as well as those more tion provided by the high spectral resolution of the [C II] generally predicted through Starburst99 simulations. data that the radio data lacks, we were successful in de- We do not propose any solutions to this conundrum, coupling the central “ridge” from the shell. The “radio at present; this requires further detailed analyses of the ring B” surrounding WR20b appears to be a superimpo- member stars, especially WR20a and 20b, as well as the sition of filamentary structures, rather than a coherent expanding shell. ring. These structures generally extend toward the main Wd2 cluster, suggesting that Wd2, rather than WR20b, RCW 49 21 dominantly influences their morphology. We find that that a fraction of it constitutes triggered star forma- the ridge, too, is a superimposition of hot gas and dust tion in the shell. Follow-up studies in X-ray and IR components well-separated in velocity space, which may wavelengths can shed more light on the triggered star explain the variation in H137β and H109α radio recom- formation efficiency of the shell. 4 bination line velocities observed by Benaglia et al.(2013) The stellar mass of the Wd2 cluster is ∼ 3.1 × 10 M 4 and Paladini et al.(2015), respectively. Some red-shifted for stars with masses < 0.65 M and ∼ 4 × 10 M sections of the ridge seem to be connected in velocity for stars with masses > 0.65 M (Zeidler et al. 2017). space to a larger +16 km/s molecular cloud (Furukawa This means that the stellar mass due to triggered star et al. 2009), which may indicate that these sections lie formation in the shell is smaller than the mass of Wd2 beyond the cluster and limb-brightened shell. cluster. Furthermore, from their modeling results that While we argue that the expansion of the shell is calculate emission from envelopes, disks, and outflows driven by stellar winds of WR20a, on a larger scale the surrounding stars, Whitney et al.(2004) found the most observed velocity structure (blue and red-shifted compo- massive YSO in RCW 49 is < 5.9 M , suggesting that nents of gas) in RCW 49 could be guided by the dynam- the new generation of stars will be relatively lower in ics of several individual molecular clouds which predate mass compared to the stars in Wd2 and feedback from Wd2. Furukawa et al.(2009) suggests that a collision this next generation of stars is expected to be limited. between two of these clouds may have contributed to the formation of Wd2 and, consequently, RCW 49. Ad- 4.3. Comparison with the shell of Orion ditionally, Townsley et al.(2019) observed diffuse hard The Orion Molecular Cloud (OMC) is the closest mas- X-ray emission from far-west of Wd2 toward a pulsar sive star-forming region that has been studied exten- wind nebula that is indicative of a cavity supernova rem- sively in a wide range of wavelengths. OMC 1 is its nant, suggesting an earlier generation of massive star most massive core associated with the well known HII formation in RCW 49. Perhaps this earlier generation region of M42 (the Orion Nebula). Pabst et al.(2019) of star formation is responsible for the large scale veloc- reported an expanding shell driven by the stellar winds ity dispersion observed by Furukawa et al.(2009), while of the O7V star θ1 Ori C in the Orion Nebula. The the local shell expansion that we present in this work is velocity of the expanding Orion veil shell is similar to driven by the stellar winds of WR20a. that of the shell of RCW 49, i.e. 13 km s−1, but has a Whitney et al.(2004) studied star formation in dif- mass of ∼ 2600 M , which is about 9 times lower than ferent regions of RCW 49 using the GLIMPSE survey the mass (and also the kinetic energy) of the shell of and found that most of the star formation is occurring RCW 49. The difference between the kinetic energies within a 5 pc radius (similar to the transition bound- reflects the fact that the Orion veil shell is the result of ary of Churchwell et al. 2004) from the Wd2 cluster. the mechanical energy input from one O7V star while At larger distances, a second generation of star forma- the shell in RCW 49 is the effect of a rich stellar cluster. tion perhaps triggered by Wd2 is also suggested, based Furthermore, Orion with an age of 0.2 Myrs is relatively on the massive (B2–3) young stellar objects (YSOs) de- younger compared to RCW 49, that has an age of at tected. To put it in context of our findings, this implies least 2 Myrs. Despite being created by a larger mechan- that star formation is occurring mainly in the ridge of ical input, the shell of RCW 49 is moving at a similar RCW 49, while a second (younger) generation of star velocity as that of the Orion veil. Perhaps this is because formation is probably triggered in the shell. While the the shell of RCW 49 is broken toward the west and is former one may reflect the cloud-cloud collision event venting out plasma, while the Orion veil seems to be a highlighted by Furukawa et al.(2009), the latter is likely complete shell. However, likely, the veil will burst soon triggered by feedback from the Wd2 cluster. However, as well, releasing the hot plasma and hence the driving Hur et al.(2015) suggests triggered star formation from force of the expansion. the radiative feedback of Wd2 as a possible explana- The O7V star θ1 Ori C lies at the front-side of OMC tion for the enhanced abundance of pre-main sequence 1 and the shell expansion toward the rear is stopped by (PMS) candidates observed in the ridge in X-ray by the dense core. In contrast, in RCW 49, backside of the Naz´eet al.(2008). We suggest both the cloud-cloud col- bubble seems to have broken and the large scale molec- lision event and the compression from Wd2’s radiative ular cloud in the velocity range of 11 to 21 km s−1 (as feedback could be a cause for the triggered star forma- reported by Furukawa et al. 2009) partially blocks the tion in the ridge. expansion of the red-shifted gas. Another interesting Whitney et al.(2004) reported a total of ∼ 7000 YSOs difference between the two shells is that we observe CO in RCW 49 with a total mass of 4500 M and we infer emission toward the same line-of-sight of the RCW 49’s 22 Tiwari et al. shell and its spatial distribution, though fragmented, the surroundings, the expansion stalls but if the stars outlines the shell (as in Figs.4 and5), while the Orion are massive enough to enter the Wolf-Rayet phase, the veil shell lacks CO emission (Pabst et al. 2020). Non- expansion can be rejuvenated. detection of CO in the Orion veil shell is attributed to the rather limited column density, Av ∼ 2 mag, which corresponds to a gas column of N(H) = 4 × 1021 cm−2 5. CONCLUSIONS (Pabst et al. 2020), while we derived a maximum Av We presented for the first time large scale velocity ∼ 18 for RCW 49, which corresponds to a gas column 2 2 integrated intensity maps of P3/2 → P1/2 transition 22 −2 12 13 of N(H) = 3 × 10 cm . Existence of larger scale of [C II], J = 3 → 2 transition of CO and CO to- (and perhaps older) molecular clouds in RCW 49 has ward RCW 49. By analyzing the observed data in dif- been established (Furukawa et al. 2009). The old age ferent velocity ranges, we successfully decoupled an ex- of RCW 49 (2 Myrs) as compared to Orion (0.2 Myrs) panding shell associated with RCW 49 from the entire puts RCW 49 at an advanced stage of evolution and gas complex. With the accessibility of better resolution more pronounced effects of stellar feedback in shaping data compared to the earlier studies (as discussed in its environment by sweeping dense molecular clouds seen Sect. 4.2) done toward RCW 49, we justified the pres- as clumps toward the shell. Moreover, despite having a ence of a single shell instead of previously thought two larger mechanical input, the denser gas column environ- and characterised it for the first time. We find that the ment of RCW 49 could be another reason for its shell’s shell expanding toward us at ∼ 13 km s−1 is ∼ 1 pc expansion velocity to be similar to that of the Orion thick and has a radius of ∼ 6 pc. We used dust SEDs 13 veil. The larger statistical study of the effects of stellar and the column densities of [C II] and CO, to estimate feedback in Galactic star forming regions initiated by 4 the mass of the shell ∼ 2.5 × 10 M . We quantified the SOFIA FEEDBACK legacy program can help illu- and discussed the effects of the stellar wind feedback, minate whether swept shells typically resemble Orion or which mechanically powers the expansion of the shell of RCW 49. RCW 49. We constrained the physical conditions of the hot plasma and the ionised gas using the previous X-ray 4.4. Our understanding of stellar feedback and radio wavelength studies, while using our new [C II] This study of the expanding shell and molecular and CO observations to determine the PDR parameters. clouds in RCW 49 contributes toward our understand- Building on the geometry of RCW 49 derived from the ing of the stellar feedback in our Galaxy. In Sect. 3.6.2 dust emission studies, we put forward a 3D represen- tation of the shell as seen by the observer, where the we find that the evolution of the hot plasma, the H II re- gion and the PDR seems to be dominated by the energy [C II] shell overhangs the transition boundary between injection by stellar winds of massive stars. the ionised gas and the PDR. Based on the energy and Following our discussion on the next generation of star time scale estimations, we suggest that the shell, ini- formation in Sect. 4.2, the total mass available for (trig- tially powered by Wd2, broke open in the west releasing 4 the hot plasma and its observed re-acceleration is mainly gered) star formation in the swept up shell is ∼ 10 M , which can be compared to the molecular clouds (of total driven by the Wolf-Rayet star, WR20a. 5 Besides the qualitative and quantitative analysis of mass ∼ 2 × 10 M ) from which Wd2 was formed (Fu- rukawa et al. 2009). As, even in a dense core, the star the shell, we spectrally resolve and present the spatially formation efficiency is less than unity, the total mass of distinct gas structures in RCW 49: the ridge and the the triggered cluster will be considerably less than that northern and southern clouds. of Wd2. Moreover, as the mass of the most massive star Comparing our findings with the existing literature, in a cluster scales with the mass of the cluster (McKee we conclude that a secondary generation of star for- & Williams 1997) and feedback can be expected to scale mation has been triggered in the shell but the new with the mechanical luminosity injected by the (most generation of stars being formed are relatively lower in massive) star. Therefore, because each successive trig- mass than those existing in Wd2. gered star cluster will have lower mass than the previous, resulting in lower feedback, the triggered star formation process will gradually decrease. Furthermore, comparing RCW 49 with Orion, we sur- ACKNOWLEDGMENTS mise that the effects of the stellar winds are limited to the earliest phases of the expansion and that, once the We thank the anonymous referee for bringing impor- swept up shell breaks open, the hot gas is vented into tant issues to our attention and for helping to clarify the paper. RCW 49 23

This work is based on observations made with the The FEEDBACK project is supported by the Federal NASA/DLR Stratospheric Observatory for Infrared As- Ministry of Economics and Energy (BMWI) via DLR, tronomy (SOFIA). SOFIA is jointly operated by the Projekt Number 50 OR 1916 (FEEDBACK) and Pro- Universities Space Research Association, Inc. (USRA), jekt Number 50 OR 1714 (MOBS - MOdellierung von under NASA contract NNA17BF53C, and the Deutsches Beobachtungsdaten SOFIA). SOFIA Institut (DSI) under DLR contract 50 OK 0901 This work was also supported by the Agence to the University of Stuttgart. Financial support for the National de Recherche (ANR/France) and the SOFIA Legacy Program, FEEDBACK, at the Univer- Deutsche Forschungsgemeinschaft (DFG/Germany) sity of Maryland was provided by NASA through award through the project “GENESIS” (ANR-16-CE92-0035- SOF070077 issued by USRA. 01/DFG1591/2-1).

APPENDIX

A. IMPROVED BASELINE REDUCTION ALGORITHM USING A PRINCIPAL COMPONENT ANALYSIS A general description of PCA technique and its application in astrophysics can be found in Heyer & Peter Schloerb (1997) and Ungerechts et al.(1997). We employ this novel method to produce upGREAT spectra with a higher quality than is possible with a standard polynomial baseline removal. To do so we identify systematic variations of the baseline between multiple spectra of the same receiver element. These variations are caused by instabilities in the telescope system, i.e. backends, receiver, telescope (optics), and atmosphere over the course of the observations. In the calibration step we produce additional data from the OFF-source (emission-free background) measurements of the on-the-fly data by subtracting subsequent OFF positions from each other. We calibrate these “OFF-OFF” spectra in the same way the “ON-OFF” spectra are calibrated. This additional set of spectra contains all the dynamics of the telescope system and the atmosphere without the astronomical information. By the use of a Principal Component Analysis (PCA) we identify systematic “components” or “eigenspectra” that explain most of the variance away from the mean between these spectra. This is done separately for each of the 14 receiver elements of the upGREAT array and for each flight in which the source was observed. Using a linear combination of the strongest “components” we try to describe the “ON-OFF” spectra as best as possible by finding the best fit coefficients for each component. Subsequently we scale each component by the coefficients that we found and subtract them from the “ON-OFF” spectra. This removes the systematic variations found in the “OFF-OFF” spectra, but does not alter the astronomical information in the “ON-OFF” spectra. With this technique we can correct very complex baseline features that are difficult or impossible to correct with the standard polynomial baseline removal. We are currently preparing a paper that describes this method in more detail (Buchbender et al. in prep.). 24 Tiwari et al.

B. EXPANSION OF THE SHELL

−1 −1 Figure 14. Different channels of [C II] emission in the velocity range of -14 to -12 km s (red) and -4 to -2 km s (blue), depicting the expansion of the shell as seen in Fig.3. RCW 49 25

C. PV DIAGRAMS

−1 12 Figure 15. (Left) Integrated [C II] intensity from −10 to 0 km s in greyscale. Integrated CO(3−2) within the same velocity interval in contours, which mark integrated intensities of 20, 50, and 125 K km s−1 in black, blue, and red contours respectively. The overlaid red curve tracing the limb-brightened shell marks the path for the pv diagram in the right hand plot. (Right) The pv diagram along the red path in the left hand plot, with 00 displacement at the southern end of the red curve. As in the right 12 hand plot, [C II] in gray-scale and CO(3−2) in contours, with black, blue, and red contours marking intensities of 4, 8, and 16 K respectively. In both the integrated intensity and pv diagram, 13CO(3−2) generally follows the 12CO. 26 Tiwari et al.

−1 Figure 16. The first two columns show the [C II] velocity integrated intensity maps in the range from −25 to 0 km s and from 0 to 30 km s−1, respectively. They are marked with vertical and horizontal cuts (white dashed lines), along which the pv diagrams are shown in columns named vertical and horizontal, respectively. The color bar is common for all panels. RCW 49 27

D. OPTICAL DEPTH OF [C II] 13 13 We determined the opacity of [C II] emission using its isotope [ C II] . The [ C II] line splits into three hyperfine- structure (hfs) components due to the coupling of angular momentum and spin of the hydrogen nucleus. Due to the limited signal-to-noise (S/N) ratio we have toward any single line-of-sight, we averaged the spectra toward a bright 00 00 00 00 13 region (100 × 100 around ∆α = 150 , ∆δ = 80 ) in [C II], emission, to detect [ C II] lines. We used the F = 1 → 0 hfs component at 1900.95 GHz, which is the second strongest hfs component with a relative intensity (ri) of 0.25 13 (see Guevara et al. 2020, Table. 1), to calculate the total [ C II] intensity. The reason we did not use the brightest hfs component is because it lies within the velocity wing of the [C II] emission, while the second strongest component 13 lies well beyond the velocity range of [C II] emission and can be analysed. The F = 1 → 0 hfs component of [ C II] is multiplied by the 12C/13C ratio, α = 52 (Milam et al. 2005), for a Galactocentric distance of ∼ 8 kpc for RCW 49 (using equation 2 of Brand & Blitz 1993).

13 As can be seen in Fig. 17, we find that the [ C II] spectral emission follows a similar profile as that of [C II] within its higher noise, but the intensity is higher than expected for optically thin [C II] , indicative of an optical depth > 1. Using the technique mentioned in Guevara et al.(2020) and using their equation 4, we estimated an optical depth in [C II] , of τ = 3. This number should only be taken as a reference because we find that for an rms of 0.8 K, we get a 13 [ C II] peak detection of ∼ 1.2 K i.e. a S/N of 1.5. Due to reduced S/N, we were unable to estimate [C II] optical depths toward other regions. So, we take τ = 3 as an upper limit for the entire [C II] emission toward RCW 49.

13 00 00 00 00 Figure 17. Average spectra of [C II] and [ C II] toward a bright 100 × 100 area around (∆α = 150 , ∆δ = 80 ). The 12 13 13 C/ C ratio is denoted by α and ri is the relative intensity of F = 1 → 0 hfs component of [ C II].

E. CO EXCITATION TEMPERATURE AND COLUMN DENSITY Assuming 12CO to be optically thick and that both 12CO and 13CO have same excitation conditions under LTE, we 12 can adapt the formalism as described in Tiwari et al.(2018) to determine Tex. Using J = 3 → 2 transition of CO, we can calculate Tex by

  16.6 −1 Tex = 16.6 ln 1 + 12 K , (E1) TMB( CO) where Tmb is the main beam brightness temperature and the constant (hν/k = 16.6 K) is calculated for ν = 12 12 345.769 GHz, which is the frequency for J = 3 → 2 transition of CO. For instance, for an average TMB( CO) within the masked region shown in Fig. 11 is ∼ 8 K, we get an average Tex ∼ 15 K. Using the pixel-by-pixel calculation 13 of Tex, we can determine the N( CO) using

! ! Z 13 12 31.7 τ13 13 −2 N( CO) = 5.29 × 10 (Tex + 0.88) exp TMB( CO)dv cm . (E2) Tex 1 − exp(−τ13) 28 Tiwari et al.

3 2 12 Here the constants are 3kQrot/8π νµ Jup = 5.29 × 10 (Tex + 0.88) and the upper level energy, Eup = 31.7 K, −19 determined for the partition function, Qrot = 0.38Tex + 1/3, the dipole moment, µ = 1.1 × 10 esu and the upper 13 12 level, Jup = 3. The optical depth of CO, τ13, can be calculated for an optically thick CO by:

" 13 −1# TMB( CO) 1  τ13 = −ln 1 − − 0.003 . (E3) 15.87 exp(15.87/Tex) − 1

Here the constants are hν/k = 15.873 K and 1/(exp(hν/kTbg) - 1) = 0.003, calculated for ν = 330.588 GHz, 13 which is the frequency for J = 3 → 2 transition of CO and for a background temperature, Tbg = 2.75 K. For the column density estimation within the half-elliptical mask shown in Fig. 11, we used the average main beam brightness 12 13 temperatures of CO and CO and determined an average τ13 = 0.325.

F. WESTERLUND 2 CATALOG AND STELLAR PROPERTIES In order to estimate the potential influence of the known stellar population on its environment, we synthesized a catalog of early-type stars from the work of Tsujimoto et al. 2007 (hereafter TFT07), Vargas Alvarez´ et al. 2013 (hereafter VA13), and Mohr-Smith et al. 2015 (hereafter VPHAS+). From these three catalogs, we should collect most of the known O and early B stars associated with Westerlund 2. We intended to collect as many early-type candidates as possible so that we could evaluate the feedback contribution of the few most massive stars compared to the contribution of the full catalog of suspected OB stars, so we accepted stars which fulfilled any of several relaxed constraints. We admitted into our list of early-type candidates any stars 1) with known OB spectral types listed by VA13 in their Table 6 or VPHAS+; or 2) marked by VPHAS+ as ‘WD2’ cluster candidates based on similar extinction values; or 3) marked by TFT07 as ‘ET’ candidates based on their NIR colors; or 4) present in Table 3 of VA13 in sub-tables “Stars with only absorption lines” or “Stars believed to be late O/early B.” In order to cast a wide net for OB candidates, we only required stars to fulfill one of these criteria, though many fulfilled several. VPHAS+ identified objects in their catalog that were also cataloged by TFT07 or VA13, and we did no further cross-matching of our own between the VPHAS+ catalog and either of the other two. For candidates from either TFT07 or VA13 that were not in the VPHAS+ catalog, we cross-matched between those two catalogs. We find a total of 83 massive stars associated with Wd2, though these include a few stars that VPHAS+ identifies as potential runaways or ejectees due to their similar reddening to the rest of the cluster. Of the 83 total massive stars, 66 are within 120 of the cluster center, close enough to influence the thermal and kinematic properties of the 0 0 H II region; 60 are within 6 , and 50 are within 3 . We did not explicitly check our synthesized catalog against those of Rauw et al.(2011) or Hur et al.(2015), among others, so it’s possible that we could be missing a small number (∼< 5) of O or early B stars within ∼12’. We are aware that Zeidler et al.(2018), by their analysis of VLT /MUSE spectra, add 2 new O stars (O7.5 and O8.5) and 5 new B stars to the list of known spectral types near the cluster center, which we have not included in our catalog. We adopted OB spectral types first from VPHAS+, since it is the most recent work, and then from VA13. For all remaining stars, we assigned the uncertain type O8–B1.5.

We used the theoretical calibrations of Martins et al.(2005) to estimate effective temperature Teff, surface gravity log g, and luminosity L from spectral type. For WR20a, we adopt the fitted parameters Teff, R∗, v∞, and M˙ from Rauw et al.(2005), who has suggested that the two binary components are nearly identical based on their spectra. We note a potential caveat here: Rauw et al.(2005) assumes a heliocentric distance close to 8 kpc. We use only the four spectroscopically fitted parameters listed above, and it is not clear to us how the heliocentric distance affects those particular parameters in their analysis. In any case, this is the only available measurement of these parameters which are necessary for our feedback capacity analysis of the WR stars. While the spectrum of WR20b has not been modeled in such detail, the star has been assigned the same type (WN6ha; van der Hucht 2001) as each component of WR20a (WN6ha+WN6ha; Rauw et al. 2005), so we adopt the same parameters for WR20b as for one component of WR20a. Given the appropriate parameters, we picked out models for each star from the PoWR stellar atmosphere grids (Sander et al. 2015). Each of the PoWR models provides a synthetic spectrum, from which we integrate the total ionizing flux between 6–13.6 eV in order to calculate G0 from the ensemble of stars. For each location in a coordinate grid covering the entire H II region, we add up the ionizing flux from every star using projected distances assuming all stars lie in a RCW 49 29 plane at 4.16 kpc. We can use this grid to find G0 at any location throughout the region; at the location of the bright 3 Eastern shell, G0 ∼ 2–3 × 10 in Habing units. For the total mass loss rate M˙ of the cluster, we take the individual mass loss rates of the OB stars from Leitherer et al.(2010) using the Teff and log g calculated above. For the WR stars, we use the mass loss rate from Rauw et al. (2005). We sum over all stars in the cluster to find the total mass loss rate. Finally, we calculate the total mechanical luminosity Lmech of the cluster by summing over Lmech of each star. We 1 ˙ 2 ˙ calculate Lmech = 2 Mv∞ using the terminal wind velocities v∞ and mass loss rates M of each star from Leitherer et al.(2010) for OB and Rauw et al.(2005) for WR.

We take a statistical approach to determining the values and uncertainties of these cluster properties. For each star, we create a set of possible spectral types including 1) an inherent half sub-type calibration uncertainty and 2) the stated range of possible spectral types, when present. For the WR stars, we take stated uncertainties associated with the parameters assigned by Rauw et al.(2005) and sample from them random realizations of parameter combinations ˙ for the WR stars. Uncertainty was not specified for mass loss rate, so we assume√ a 10% inherent uncertainty in M, though we primarily drive the mass loss rate uncertainty by scaling it with f, where f is the volume filling factor which we vary between 0.1–0.25 (Rauw et al. 2005 assumes 0.1, while Todt et al. 2015 assumes 0.25). With these sets of possible spectral types or parameter combinations for each cluster member, we draw random realizations of the entire cluster. For each realization, we assign FUV flux, mass loss rate, and mechanical luminosity as described above, sum them across the entire cluster realization, and then take the median of the summed values of all cluster realizations. We use the 16th and 84th percentiles of these distributions for the lower and upper error bars, respectively.

G. COMPARISON TO STARBURST99 We augment our analysis of the energetics in Section 3.6.2 with predictions made using Starburst99, a spectral synthesis software which accepts a cluster IMF description and simulates population synthesis and cluster evolution over time by following stellar evolution models (Leitherer et al. 2014). The software outputs stellar spectra, wind properties (which are of particular interest to us), and other synthesized cluster properties at desired evolutionary time steps. We can compare our configuration with that of Rauw et al.(2007), who used Starburst99 to estimate the wind power of Westerlund 2 using a Salpeter IMF and a total mass of 4500 M in stars of masses between 1 and 120 M . All inputs to the software are left as default unless otherwise specified. We base the cluster mass function properties on the IMF fitted by Ascenso et al.(2007), with a slope of Γ = −1.20±0.16 and total mass 2809 M in stars of masses between 0.8 and 11 M . From these values, we calculate the total mass in stars between 1 and 100 M . We adopt the lower limit of 1 M from Rauw et al.(2007) and the upper limit of 100 M by averaging the 80 M upper limit from the most massive observed star (Zeidler et al. 2017) and the ∼ 120 M upper limit suggested by Ascenso et al. (2007) and used by Rauw et al.(2007). The uncertainty on the IMF slope has a significant impact on the derived total cluster mass in our adopted mass bin, so we estimate the uncertainty on the modeled cluster wind properties due to the IMF slope uncertainty. We sample a large number (N ∼ 1000) of IMF slope values from the distribution suggested by Ascenso et al.(2007), Γ = −1.20 ± 0.16, assuming they describe a Gaussian distribution with µ = −1.20 and σ = 0.16, and use each value to independently calculate the total mass in stars between 1 and 100 M . From this mass distribution, we take the median (4000 M ) and 16th and 84th percentile values (2900 and 5700 M ) as the value and lower and upper error bars, respectively. In order to avoid running a large number of Starburst99 simulations, we simply run three simulations with these three total mass values, and all other inputs kept as described above. For all predictions made using Starburst99, we thus use the “median simluation” as the predicted value and the upper and lower “error bar simulations” as the error bars, as in Figure 12. We expect the total cluster mass to have a larger impact on the predicted wind power output than the IMF slope, so we do not vary the IMF slope in the simulations themselves.

REFERENCES

Arnaud, K. A. 1996, in Astronomical Society of the Pacific Ascenso, J., Alves, J., Beletsky, Y., & Lago, M. T. V. T. Conference Series, Vol. 101, Astronomical Data Analysis 2007, A&A, 466, 137, doi: 10.1051/0004-6361:20066433 Software and Systems V, ed. G. H. Jacoby & J. Barnes, 17 30 Tiwari et al.

Benaglia, P., Koribalski, B., Peri, C. S., et al. 2013, A&A, Kaufman, M. J., Wolfire, M. G., & Hollenbach, D. J. 2006, 559, A31, doi: 10.1051/0004-6361/201321976 ApJ, 644, 283, doi: 10.1086/503596 Benjamin, R. A., Churchwell, E., Babler, B. L., et al. 2003, Klein, B., Hochg¨urtel,S., Kr¨amer,I., et al. 2012, A&A, PASP, 115, 953, doi: 10.1086/376696 542, L3, doi: 10.1051/0004-6361/201218864 Bennett, C. L., Fixsen, D. J., Hinshaw, G., et al. 1994, Leitherer, C., Ekstr¨om,S., Meynet, G., et al. 2014, ApJS, ApJ, 434, 587, doi: 10.1086/174761 212, 14, doi: 10.1088/0067-0049/212/1/14 Bohlin, R. C., Savage, B. D., & Drake, J. F. 1978, ApJ, Leitherer, C., Ortiz Ot´alvaro, P. A., Bresolin, F., et al. 224, 132, doi: 10.1086/156357 2010, ApJS, 189, 309, doi: 10.1088/0067-0049/189/2/309 Brand, J., & Blitz, L. 1993, A&A, 275, 67 Lombardi, M., Bouy, H., Alves, J., & Lada, C. J. 2014, Butterfield, N. 2018, PhD thesis, The University of Iowa A&A, 566, A45, doi: 10.1051/0004-6361/201323293 Cantat-Gaudin, T., Jordi, C., Vallenari, A., et al. 2018, Martins, F., Schaerer, D., & Hillier, D. J. 2005, A&A, 436, A&A, 618, A93, doi: 10.1051/0004-6361/201833476 1049, doi: 10.1051/0004-6361:20042386 Castellanos, P., Bern´e,O., Sheffer, Y., Wolfire, M. G., & McKee, C. F., & Williams, J. P. 1997, ApJ, 476, 144, Tielens, A. G. G. M. 2014, ApJ, 794, 83, doi: 10.1086/303587 doi: 10.1088/0004-637X/794/1/83 Milam, S. N., Savage, C., Brewster, M. A., Ziurys, L. M., & Churchwell, E., Whitney, B. A., Babler, B. L., et al. 2004, Wyckoff, S. 2005, ApJ, 634, 1126, doi: 10.1086/497123 Mohr-Smith, M., Drew, J. E., Barentsen, G., et al. 2015, ApJS, 154, 322, doi: 10.1086/422504 MNRAS, 450, 3855, doi: 10.1093/mnras/stv843 Churchwell, E., Povich, M. S., Allen, D., et al. 2006, ApJ, Molinari, S., Swinyard, B., Bally, J., et al. 2010, PASP, 649, 759, doi: 10.1086/507015 122, 314, doi: 10.1086/651314 Cowie, L. L., & McKee, C. F. 1977, ApJ, 211, 135, Naz´e,Y., Rauw, G., & Manfroid, J. 2008, A&A, 483, 171, doi: 10.1086/154911 doi: 10.1051/0004-6361:20078851 Dalgarno, A., & McCray, R. A. 1972, ARA&A, 10, 375, Pabst, C., Higgins, R., Goicoechea, J. R., et al. 2019, doi: 10.1146/annurev.aa.10.090172.002111 Nature, 565, 618, doi: 10.1038/s41586-018-0844-1 Draine, B. T. 2003, ARA&A, 41, 241, Pabst, C. H. M., Goicoechea, J. R., Teyssier, D., et al. doi: 10.1146/annurev.astro.41.011802.094840 2020, A&A, 639, A2, doi: 10.1051/0004-6361/202037560 Drew, J. E., Herrero, A., Mohr-Smith, M., et al. 2018, Paladini, R., Ingallinera, A., Agliozzo, C., et al. 2015, ApJ, MNRAS, 480, 2109, doi: 10.1093/mnras/sty1905 813, 24, doi: 10.1088/0004-637X/813/1/24 Elmegreen, B. G., & Lada, C. J. 1977, ApJ, 214, 725, Pilbratt, G. L., Riedinger, J. R., Passvogel, T., et al. 2010, doi: 10.1086/155302 A&A, 518, L1, doi: 10.1051/0004-6361/201014759 Furukawa, N., Dawson, J. R., Ohama, A., et al. 2009, Planck Collaboration, Aghanim, N., Ashdown, M., et al. ApJL, 696, L115, doi: 10.1088/0004-637X/696/2/L115 2016, A&A, 596, A109, Garmire, G. P., Bautz, M. W., Ford, P. G., Nousek, J. A., doi: 10.1051/0004-6361/201629022 & Ricker, George R., J. 2003, in Society of Photo-Optical Poglitsch, A., Waelkens, C., Geis, N., et al. 2010, A&A, Instrumentation Engineers (SPIE) Conference Series, 518, L2, doi: 10.1051/0004-6361/201014535 Vol. 4851, X-Ray and Gamma-Ray Telescopes and Pound, M. W., & Wolfire, M. G. 2008, in Astronomical Instruments for Astronomy., ed. J. E. Truemper & H. D. Society of the Pacific Conference Series, Vol. 394, Tananbaum, 28–44, doi: 10.1117/12.461599 Astronomical Data Analysis Software and Systems XVII, Guevara, C., Stutzki, J., Ossenkopf-Okada, V., et al. 2020, ed. R. W. Argyle, P. S. Bunclark, & J. R. Lewis, 654 arXiv e-prints, arXiv:2002.12692. Rauw, G., Manfroid, J., Gosset, E., et al. 2007, A&A, 463, https://arxiv.org/abs/2002.12692 981, doi: 10.1051/0004-6361:20066495 G¨usten,R., Nyman, L. A.,˚ Schilke, P., et al. 2006, A&A, Rauw, G., Sana, H., & Naz´e,Y. 2011, A&A, 535, A40, 454, L13, doi: 10.1051/0004-6361:20065420 doi: 10.1051/0004-6361/201117000 Heyer, M. H., & Peter Schloerb, F. 1997, ApJ, 475, 173, Rauw, G., Crowther, P. A., De Becker, M., et al. 2005, doi: 10.1086/303514 A&A, 432, 985, doi: 10.1051/0004-6361:20042136 Hollenbach, D. J., & Tielens, A. G. G. M. 1999, Reviews of Risacher, C., G¨usten, R., Stutzki, J., et al. 2018, Journal of Modern Physics, 71, 173, Astronomical Instrumentation, 7, 1840014, doi: 10.1103/RevModPhys.71.173 doi: 10.1142/S2251171718400147 Hur, H., Park, B.-G., Sung, H., et al. 2015, MNRAS, 446, Sander, A., Shenar, T., Hainich, R., et al. 2015, A&A, 577, 3797, doi: 10.1093/mnras/stu2329 A13, doi: 10.1051/0004-6361/201425356 RCW 49 31

Schneider, N., Simon, R., Guevara, C., et al. 2020, PASP, van der Hucht, K. A. 2001, NewAR, 45, 135, 132, 104301, doi: 10.1088/1538-3873/aba840 doi: 10.1016/S1387-6473(00)00112-3 Schuller, F., Menten, K. M., Contreras, Y., et al. 2009, Vargas Alvarez,´ C. A., Kobulnicky, H. A., Bradley, D. R., A&A, 504, 415, doi: 10.1051/0004-6361/200811568 et al. 2013, AJ, 145, 125, Sofia, U. J., Lauroesch, J. T., Meyer, D. M., & Cartledge, doi: 10.1088/0004-6256/145/5/125 S. I. B. 2004, ApJ, 605, 272, doi: 10.1086/382592 Wareing, C. J., Pittard, J. M., Wright, N. J., & Falle, S. A. E. G. 2018, MNRAS, 475, 3598, Stacey, G. J., Geis, N., Genzel, R., et al. 1991, ApJ, 373, doi: 10.1093/mnras/sty148 423, doi: 10.1086/170062 Weaver, R., McCray, R., Castor, J., Shapiro, P., & Moore, Tielens, A. G. G. M. 2008, ARA&A, 46, 289, R. 1977, ApJ, 218, 377, doi: 10.1086/155692 doi: 10.1146/annurev.astro.46.060407.145211 Whiteoak, J. B. Z., & Uchida, K. I. 1997, A&A, 317, 563 —. 2010, The Physics and Chemistry of the Interstellar Whitney, B. A., Indebetouw, R., Babler, B. L., et al. 2004, Medium ApJS, 154, 315, doi: 10.1086/422557 Tiwari, M., Menten, K. M., Wyrowski, F., et al. 2018, Wiesenfeld, L., & Goldsmith, P. F. 2014, ApJ, 780, 183, A&A, 615, A158, doi: 10.1051/0004-6361/201732437 doi: 10.1088/0004-637X/780/2/183 Todt, H., Sander, A., Hainich, R., et al. 2015, A&A, 579, Williams, J. P., & McKee, C. F. 1997, ApJ, 476, 166, A75, doi: 10.1051/0004-6361/201526253 doi: 10.1086/303588 Townsley, L. K., Broos, P. S., Garmire, G. P., & Povich, Young, E. T., Becklin, E. E., Marcum, P. M., et al. 2012, M. S. 2019, ApJS, 244, 28, ApJL, 749, L17, doi: 10.1088/2041-8205/749/2/L17 doi: 10.3847/1538-4365/ab345b Zavagno, A., Anderson, L. D., Russeil, D., et al. 2010, Townsley, L. K., Feigelson, E. D., Montmerle, T., et al. A&A, 518, L101, doi: 10.1051/0004-6361/201014587 2003, ApJ, 593, 874, doi: 10.1086/376692 Zeidler, P., Nota, A., Grebel, E. K., et al. 2017, AJ, 153, 122, doi: 10.3847/1538-3881/153/3/122 Tsujimoto, M., Feigelson, E. D., Townsley, L. K., et al. Zeidler, P., Sabbi, E., Nota, A., et al. 2015, AJ, 150, 78, 2007, ApJ, 665, 719, doi: 10.1086/519681 doi: 10.1088/0004-6256/150/3/78 Ungerechts, H., Bergin, E. A., Goldsmith, P. F., et al. 1997, —. 2018, AJ, 156, 211, doi: 10.3847/1538-3881/aae258 ApJ, 482, 245, doi: 10.1086/304110